Sequential bayesian prediction in the presence of changepoints and faults

Journal article

We introduce a new sequential algorithm for making robust predictions in the presence of changepoints. Unlike previous approaches, which focus on the problem of detecting and locating changepoints, our algorithm focuses on the problem of making predictions even when such changes might be present. We introduce nonstationary covariance functions to be used in Gaussian process prediction that model such changes, and then proceed to demonstrate how to effectively manage the hyperparameters associated with those covariance functions. We further introduce covariance functions to be used in situations where our observation model undergoes changes, as is the case for sensor faults. By using Bayesian quadrature, we can integrate out the hyperparameters, allowing us to calculate the full marginal predictive distribution. Furthermore, if desired, the posterior distribution over putative changepoint locations can be calculated as a natural byproduct of our prediction algorithm. © 2010 The Author. Published by Oxford University Press on behalf of The British Computer Society. All rights reserved.

Authors: Garnett, R; Osborne, M; Reece, S; Rogers, A; Roberts, S

• Journal: Computer Journal
• Volume: 53
• Issue: 9
• Pages: 1430-1446
• Publication date: 2010-11-01
• DOI: 10.1093/comjnl/bxq003
• EISSN: 1460-2067
• ISSN: 0010-4620
• Language: English
• Keywords: changepoint detection; fault detection; Bayesian methods; time-series prediction; Gaussian processes